Report (problems with) DEAD-at-start cache_peers

[rousskov]

When a cache_peer is (re)configured, Squid probes its address using a throw-away TCP connection. If and only if that TCP probe succeeds, peer selection algorithms may pick the peer. Until then, the cache_peer is treated as if it was DEAD.

This change preserves the logic summarized above but starts reporting the initial probe failure and the fact that it marks the cache_peer DEAD. Without these reports, admins often naturally assume that the cache_peer is alive, especially if traffic can be served without it.